東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁 到查詢結果 [ null ]

切換: 標籤 | MARC模式 | ISBD

Optimization and Statistical Inferen...

Niu, Xiaochun,

FindBook

Google Book

Amazon

博客來

Optimization and Statistical Inference in Networks /

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Optimization and Statistical Inference in Networks // Xiaochun Niu.
作者:	Niu, Xiaochun,
面頁冊數:	1 electronic resource (268 pages)
附註:	Source: Dissertations Abstracts International, Volume: 86-02, Section: B.
Contained By:	Dissertations Abstracts International86-02B.
標題:	Statistics. -
電子資源:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31489624
ISBN:	9798384016557

Optimization and Statistical Inference in Networks /
Niu, Xiaochun,

Optimization and Statistical Inference in Networks /Xiaochun Niu. - 1 electronic resource (268 pages)

Source: Dissertations Abstracts International, Volume: 86-02, Section: B.

This dissertation studies optimization and statistical inference within large networks, focusing on distributed optimization and community detection problems. We aim to improve algorithm efficiency and deepen our understanding of complex networks.Distributed optimization problems aim to collaboratively optimize objectives across multi-agent systems, where significant challenges arise due to system heterogeneity. Existing distributed methods often overlook the heterogeneity in agents' computational capabilities, limiting their effectiveness. To address this, Chapters 2 and 3 introduce distributed hybrid methods that utilize these heterogeneous capabilities. We begin with a federated network, proposing a hybrid method that allows agents to perform updates based on their computational capabilities, either gradient-type or Newton-type. This method enables agents with powerful hardware to perform local Newton-type updates, thereby accelerating the overall system performance. In addition, we extend these hybrid methods to general graph topologies, tackling distributed consensus and network flow problems through different distributed approximations of Newton's methods. We prove the linear convergence rates of these methods. Numerical studies indicate that increasing the number of agents performing Newton-type updates can accelerate the overall convergence, demonstrating the method's efficacy.Community recovery is the problem of identifying latent community structures within large networks using observed data. Chapters 4 and 5 focus on exact community recovery in a family of probabilistic graph models with geometric embeddings, within the logarithmic-degree regime. These models reflect the edge-transitive properties typical of real-world networks. Our study begins with the Geometric Stochastic Block Model (GSBM), where vertices have unknown community labels and known locations in ℝd , generated by a Poisson point process. Edges are formed based on the community labels, but only between vertices within a certain distance. We identify the information-theoretic threshold for exact recovery and prove its achievability through a linear-time, two-phase algorithm: the first phase produces an almost-exact labeling via a local propagation scheme, and the second phase refines the labels. In addition, we propose the Geometric Hidden Community Model (GHCM) as an extension, including geometric versions of ℤ2- synchronization and submatrix localization as special cases. Under a mild distinctness in distributions assumption, we identify the information-theoretic threshold for this general model via a Chernoff-Hellinger divergence criterion.

English

ISBN: 9798384016557Subjects--Topical Terms:

517247
Statistics.
Subjects--Index Terms:

Chernoff-Hellinger divergence criterion

Optimization and Statistical Inference in Networks /
LDR:04158nmm a22004573i 4500 001 2430795
005 20250603111618.5
006 m o d
007 cr|nu||||||||
008 261011s2024 miu||||||m |||||||eng d
020 $a 9798384016557
035 $a (MiAaPQD)AAI31489624
035 $a AAI31489624
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Niu, Xiaochun, $e author. $0 (orcid)0000-0002-8001-989X $3 3819298
245 1 0 $a Optimization and Statistical Inference in Networks / $c Xiaochun Niu.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2024
300 $a 1 electronic resource (268 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 86-02, Section: B.
500 $a Advisors: Wei, Ermin Committee members: Gaudio, Julia; Morton, David; Nocedal, Jorge.
502 $b Ph.D. $c Northwestern University $d 2024.
520 $a This dissertation studies optimization and statistical inference within large networks, focusing on distributed optimization and community detection problems. We aim to improve algorithm efficiency and deepen our understanding of complex networks.Distributed optimization problems aim to collaboratively optimize objectives across multi-agent systems, where significant challenges arise due to system heterogeneity. Existing distributed methods often overlook the heterogeneity in agents' computational capabilities, limiting their effectiveness. To address this, Chapters 2 and 3 introduce distributed hybrid methods that utilize these heterogeneous capabilities. We begin with a federated network, proposing a hybrid method that allows agents to perform updates based on their computational capabilities, either gradient-type or Newton-type. This method enables agents with powerful hardware to perform local Newton-type updates, thereby accelerating the overall system performance. In addition, we extend these hybrid methods to general graph topologies, tackling distributed consensus and network flow problems through different distributed approximations of Newton's methods. We prove the linear convergence rates of these methods. Numerical studies indicate that increasing the number of agents performing Newton-type updates can accelerate the overall convergence, demonstrating the method's efficacy.Community recovery is the problem of identifying latent community structures within large networks using observed data. Chapters 4 and 5 focus on exact community recovery in a family of probabilistic graph models with geometric embeddings, within the logarithmic-degree regime. These models reflect the edge-transitive properties typical of real-world networks. Our study begins with the Geometric Stochastic Block Model (GSBM), where vertices have unknown community labels and known locations in ℝd , generated by a Poisson point process. Edges are formed based on the community labels, but only between vertices within a certain distance. We identify the information-theoretic threshold for exact recovery and prove its achievability through a linear-time, two-phase algorithm: the first phase produces an almost-exact labeling via a local propagation scheme, and the second phase refines the labels. In addition, we propose the Geometric Hidden Community Model (GHCM) as an extension, including geometric versions of ℤ2- synchronization and submatrix localization as special cases. Under a mild distinctness in distributions assumption, we identify the information-theoretic threshold for this general model via a Chernoff-Hellinger divergence criterion.
546 $a English
590 $a School code: 0163
650 4 $a Statistics. $3 517247
650 4 $a Applied mathematics. $3 2122814
650 4 $a Computer science. $3 523869
650 4 $a Systems science. $3 3168411
653 $a Chernoff-Hellinger divergence criterion
653 $a Geometric Stochastic Block Model
653 $a System heterogeneity
653 $a Multi-agent systems
653 $a Network flow
690 $a 0796
690 $a 0984
690 $a 0463
690 $a 0364
690 $a 0790
710 2 $a Northwestern University. $b Industrial Engineering and Management Sciences. $e degree granting institution. $3 3765828
720 1 $a Wei, Ermin $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 86-02B.
790 $a 0163
791 $a Ph.D.
792 $a 2024
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31489624